Please note, this is a STATIC archive of website hashcat.net from 08 Oct 2020, cach3.com does not collect or store any user information, there is no "phishing" involved.

hashcat Forum

Full Version: VCL over the Internet? Possible, even at loss of performance?
You're currently viewing a stripped down version of our content. View the full version with proper formatting.
Is it possible to create a cluster of computers over the Internet?
I'd like to use a few computers located at different physical locations.
For this, I only have regular Internet links (top 1 Mb upload).

What would happen? Would it work slower? How so?
If a member of the cluster is off, does another takes over it's processing?

Why is VCL so network intensive? What is exactly going through?
Everything is in wiki:

Hardware requirements
A private, high-speed, local network.
GigE is required at a minimum
Infiniband is highly recommended
No WiFi, VPN, or Internet clustering.
Hi, yes, I've read it.
The thing is what would happen if I tried? Would it lose performance or not work at all? If it were to lose performance, how much? 10%? 40%? 95%?

And anyone knows why such high bandwidth is needed?
it would be much, much slower than just cracking locally, regardless of how many gpus you had.

it's opencl middleware, so everything is sent to the compute nodes in real time. therefore you need a high bandwidth and low latency link.

if you want to cluster over the internet, you have to pre-distribute everything.
(12-07-2013, 04:50 AM)epixoip Wrote: [ -> ]it would be much, much slower than just cracking locally, regardless of how many gpus you had.

it's opencl middleware, so everything is sent to the compute nodes in real time. therefore you need a high bandwidth and low latency link.

if you want to cluster over the internet, you have to pre-distribute everything.

Yeah, like Seti@home or bitcoin.
So it COULD be done, only that VCL is not desgined to work that way, correct?
Do you know any alternative?
The only thing I can think of is cutting the dictionary / brute force attack via a central management station, that, taking into consideration how many nodes it has connected, their processing power, would distribute the works accordingly, monitor whether it receives a response from the nodes or not, sending the job to another node if one of them quits... now that I think about it, it doens't sound THAT hard to code myself...
I'd bet there is some workaround. For example, what about VPN ? You'd be able to connect all nodes to the one virtual local network. But what epixoip wrote says pretty much everything. It's about bandwidth and latency. If you could even use one of the most high-tech experimental networks with speeds around 200 Gbit/s, the latency would ruin it.